Design And Implementation Of Early Fire Warning System Based On Video Smoke Detection

Due to the high concentration of production enterprises in industrial parks,there are many fire inducements and hidden dangers of fire accidents.Therefore,great attention should be paid to fire prevention and control.Smoke is always produced before the flame,therefore smoke detection technology can realize earlier fire detection.Traditional contact smoke fire detectors are easily disturbed by dust and other particles in the environment,so they are not suitable for the wide and open fire detection environment.Video-based smoke detection method can make up for the limited detection range of traditional fire detection methods,with fast response speed and strong adaptability.It is suitable for industrial parks with large indoor space and open outdoor space.The traditional video smoke detection technology relies too much on the feature engineering,however,the color,shape and movement of the smoke are diverse,it is difficult for the feature descriptor to reflect the nature of the smoke,which affects the effect of smoke detection.Video smoke detection technology based on deep learning avoids the incomprehensiveness and blindness of manual feature extraction,but the increase of network complexity makes it difficult to implement on embedded devices.Based on the above background,this thesis researches on video smoke detection technology,and designs and develops an early fire warning system based on target detection technology and embedded technology.The main work is as follows:A video smoke detection method based on deep learning is studied for early fire prevention and control in industrial parks.In order to solve the problem of long response time of image smoke,a lightweight deep learning network is adopted to reduce the complexity of network calculation,reduce the demand of model calculation force,and improve the real-time performance of video smoke detection.In order to solve the problem of high false detection and missed detection in smoke detection,the model of smoke detection network is fine-tuned by using transfer learning,and the richness of samples is increased by using digital image enhancement technology,so as to improve the generalization ability of the model.The false detection rate and missed detection rate of video smoke detection model on the smoke sample set are effectively reduced.Based on embedded technology,the video smoke detection model is deployed to the Raspberry Pi with low cost,small size,low power consumption and high scalability,and the early fire warning system based on video smoke detection is realized.In order to solve the problem of low real-time performance of the system caused by frequency drop due to the limited computing capacity of Raspberry Pi,the sampling frequency of video frame is reduced,the calculation amount per unit time is reduced,and the real-time performance of the system is improved.After testing,the final smoke detection accuracy of the early fire warning system based on video smoke detection is 88.9%,the recall rate is 91.7%,and the reliability is high.The response time of the system to image smoke is within 15 seconds,indicating a good real-time performance.